DeepEP

DeepEP is a high-performance communication library for Mixture-of-Experts (MoE) and Expert Parallel (EP) communication.

PremiumNewProductProgrammingDeep LearningMixture-of-Experts

DeepEP is a communication library specifically designed for Mixture-of-Experts (MoE) and Expert Parallel (EP) models. It provides high-throughput and low-latency fully connected GPU kernels, supporting low-precision operations (such as FP8). The library is optimized for asymmetric domain bandwidth forwarding, making it suitable for training and inference pre-filling tasks. Furthermore, it supports Stream Multiprocessor (SM) count control and introduces a hook-based communication-computation overlap method that doesn't consume any SM resources. While its implementation differs slightly from the DeepSeek-V3 paper, DeepEP's optimized kernels and low-latency design enable excellent performance in large-scale distributed training and inference tasks.

Product Finder

Product Submit

AI Models Finder

MCP Servers

MCP Client

MCP Inspector

Case Tutorials

Latest AI News

AI Daily Brief

DeepEP

DeepEP Visit Over Time

DeepEP Visit Trend

DeepEP Visit Geography

DeepEP Traffic Sources

DeepEP Alternatives

DeepEP — DeepEP is a high-performance communication library for Mixture-of-Experts (MoE) and Expert Parallel (EP) communication.

Tencent-Hunyuan-Large — Industry-leading open-source large mixture-of-experts model

Understanding Deep Learning — Deep understanding of the principles and applications of deep learning

Moonlight-16B-A3B — Moonlight-16B-A3B is a 16B parameter Mixture-of-Experts (MoE) model trained with the Muon optimizer for efficient language generation.

Yuan2.0-M32-hf-int8 — High-Performance Mixture of Experts Language Model

Yuan2-M32-hf-int4 — High-performance mixture of experts language model

Aria — Multimodal Native Mixture of Experts Model

DeepSeek-V2.5-1210 — High-performance mixture of experts language model

DeepSeek-V3 — A Mixture-of-Experts language model with 671 billion parameters.

DeepSeek-VL2-Tiny — Advanced Large-scale Mixture of Experts Visual Language Model

DeepSeek-VL2-Small — An advanced large-scale mixture of experts visual language model.

Expert Specialized Fine-Tuning — A professional fine-tuning tool for customizing large language models.

Aria-Base-64K — Multimodal native Mixture-of-Experts model

SegMoE — SegMoE is a powerful framework that can dynamically combine Stable Diffusion models into expert mixtures within minutes, without requiring any training.

OLMoE — An open-source expert mixture language model with 130 million active parameters.

Insight Monk — A global deep technology market intelligence platform and expert community

GRIN-MoE — High-performance, low-resource consumption hybrid expert model

Moonlight — Moonlight is a 16B parameter Mixture of Experts (MoE) model trained with the Muon optimizer, delivering exceptional performance.

DualPipe — A bidirectional pipeline parallel algorithm for overlapping computation and communication in V3/R1 training.

My Expert GPT — Get your own virtual expert team, powered by your ChatGPT account.

EPLB — An open-source algorithm for expert parallelism load balancing, designed to optimize expert allocation and load balancing in multi-GPU environments.

Grok-1 — The open-sourced Grok-1 model has 314 billion parameters.

Revive AI — Learn AI through interactions with industry experts

Bytedance Flux — Flux is a fast communication overlap library for tensor/expert parallelism on GPUs.

MoE-LLaVA — An expert mixture model based on large-scale vision-language models

xinsir — Deep Learning, Representation Learning, Fine-Grained Classification

TFLearn — Advanced API simplifies TensorFlow deep learning

Fathom 2.0 — One-stop deep learning solution

AXLearn — A unified deep learning training framework.

GraphCast — Deep Learning Weather Prediction Model